MIAO: A Mental Illness Analysis Ontology for Detecting Mental Health Conditions

We introduce the Mental Illness Analysis Ontology (MIAO), an ontology designed to model the detection process of mental illness. MIAO has two key characteristics: (i) it provides a conceptual framework for the detection process, regardless of the detection method, allowing for both human and AI-driven detection approaches, and (ii) it separates the detection process from the underlying mental illness model (e.g., ontology) used to represent mental health conditions, enabling the use of different mental illness models, depending on the needs of the user or application. Two research questions guide our investigation and the development of the MIAO ontology:

• RQ₁: How can the detection process of mental illness be effectively modeled in a way that is agnostic to the specific detection method used (human or AI)?
• RQ₂: How can existing mental health models be integrated to support the detection process, ensuring compatibility across different detection systems and applications?

Ontology Development

The development of the MIAO schema follows a 4-step procedure:

1. Domain analysis

   The analysis of the domain of interest provides an overview of the state of the art, of the challenges, the strength and limitations of modelling mental illness, and especially the lack of models conceptualizing mental illness detection. The results have been reported in the previous section; this analysis also prompts the research questions (i.e., RQ₁ and RQ₂) which guide our work.

2. Specification

   After reviewing the literature and formulating the research questions, we defined a set of functional (FR_s) and non-functional (NFR_s) requirements that the ontology should satisfy. These requirements provide guidance for the initial high-level conceptualization of the ontology. Based on these requirements, we also defined a set of competency questions (CQ_s) to further refine the scope and intended use of the ontology (see REQs&CQs).

3. Construction

   The construction phase involves both the design andcformalization of the ontology. Starting from the CQs, we defined the concepts needed to model the mental illness detection domain. Also, in accordance with established best practices in ontology development, we prioritized reusing existing data models and vocabularies as a foundation, extending them to capturecdomain-specific classes and properties. Therefore, to express the concepts related to automatic detection models, we integrated ML Schema, which is used to represent and interchange information on data mining and machine learning algorithms, datasets, and experiments. Subsequently, we designed the model merging the concepts defined in our domain with those present in ML Schema. After having finalized the model we formalized the ontology in Turtle TTL format with the Protégé tool.

4. Validation

   OntOlogy Pitfall Scanner!: OOPS results

Ontology Overview

MIAO integrates:

• PROV-O – provenance modeling
• MLS – machine learning concepts
• Dublin Core Terms – metadata
• VANN – namespace management

Namespace URI: https://w3id.org/miao
Preferred Prefix: miao
Version: 1.00

Classes

Class	Description	Cardinalities / Restrictions
MentalIllnessesDetection	Represents a mental illness detection process (manual or automatic)	Must have exactly 1 input schema (usedMentalIllnessesSchema) and generates exactly 1 MentalIllnessesSet; input data (hasInputData) can be ≥1
ManualMentalIllnessesDetection	Detection performed manually by an expert	Subclass of MentalIllnessesDetection
AutomaticMentalIllnessesDetection	Detection performed automatically by a machine model	Subclass of MentalIllnessesDetection and mls:Run
MentalIllness	Detected or predicted mental illness	Belongs to exactly 1 MentalIllnessesSet; maps to exactly 1 MentalIllnessCategory (referredToMentalIllnessCategory)
MentalIllnessCategory	Category defined in a schema	Refers to ≥1 MentalIllness (referredToMentalIllness) and belongs to exactly 1 MentalIllnessesSchema
MentalIllnessesSchema	Schema or model defining categories	Must have ≥1 MentalIllnessCategory; may be used in ≥0 MentalIllnessesDetection
MentalIllnessesSet	Set of detected mental illnesses	Contains ≥1 MentalIllness; generated by exactly 1 MentalIllnessesDetection

Object Properties

Property	Domain → Range	Cardinalities / Restrictions	Description
hasMentalIllness	MentalIllnessesSet → MentalIllness	≥1 MentalIllness per set	Connects a set to its mental illnesses
belongsToMentalIllnessesSet	MentalIllness → MentalIllnessesSet	Exactly 1 set per mental illness	Inverse of hasMentalIllness
hasMentalIllnessCategory	MentalIllnessesSchema → MentalIllnessCategory	≥1 category per schema	Connects a mental health schema to its categories
isMentalIllnessCategoryOf	MentalIllnessCategory → MentalIllnessesSchema	Exactly 1 schema per category	Inverse of hasMentalIllnessCategory
referredToMentalIllnessCategory	MentalIllness → MentalIllnessCategory	Exactly 1 category per illness	Connect a mental illness to its schema category
referredToMentalIllness	MentalIllnessCategory → MentalIllness	Functional property	Inverse of referredToMentalIllnessCategory
usedMentalIllnessesSchema	MentalIllnessesDetection → MentalIllnessesSchema	Exactly 1 schema per detection	Schema used in detection
usedInMentalIllnessesDetection	MentalIllnessesSchema → MentalIllnessesDetection	≥0 detections per schema	Inverse of usedMentalIllnessesSchema
hasInputData	MentalIllnessesDetection → MLS Data	≥1 input	Input data for detection
isInputOfMentalIllnessesDetection	MLS Data → MentalIllnessesDetection	≥1 Detection	Inverse of isInputOfMentalIllnessesDetection
prov:generated	MentalIllnessesDetection → MentalIllnessesSet	Exactly 1 set generated	Output of detection
prov:wasGeneratedBy	MentalIllnessesSet → MentalIllnessesDetection	Exactly 1 detection per set	Inverse of prov:generated

Data Properties

Property	Domain	Range	Description
refersToSample	MentalIllness	string	Numerical identifier of the data sample on which one or more mental illnesses were detected
hasMentalIllnessDetectionConfidence	MentalIllness	decimal	Confidence of detection (human or ML)
hasMentalIllnessScore	MentalIllness	decimal	Numeric intensity (0–1)
hasMentalIllnessLevel	MentalIllness	string	Categorical intensity (Low, Medium, High)

Ontology Validation Case

To demonstrate the applicability of MIAO, we provide a real-world validation based on a published research study on detecting mental illnesses from social media using AI models. This case shows how MIAO can represent AI-based detection workflows, results, and evaluation metrics.

The validation resources are available in the validation-case folder.

🚀 Google Colab Notebook

We have implemented a Python Colab Notebook showing how to interact with MIAO ontology by performing SPARQL queries based on the CQ_s of each REQ (REQs&CQs).

Click below to access the interactive notebook:

⚠️ Important

To load and run the notebook correctly, please follow the instructions in the README.md.

---

Note

• For more detailed information, please refer to the paper.
• Two Turtle files showing how to model manual and automatic mental illness detection activities using MIAO are available here (also used in the notebook).
• Use Cases are available here

Authors

• Gianluca Apriceno: apriceno@fbk.eu
• Sergio Muñoz: sergio.munoz@upm.es
• Tania Bailoni: tbailoni@fbk.eu
• Mauro Dragoni: dragoni@fbk.eu
• Carlos Á. Iglesias: carlosangel.iglesias@upm.es

License

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.